Jurisdiction-enforced inferenceAccounts are reviewed per jurisdiction and billed from the first token. No free credits, no promotional tier.Contact us for a quotation
15 Sep 2026Infrastructure

Same model, different border

DeepSeek on AWS and DeepSeek on Alibaba Cloud use the same weights but carry different legal positions. Combining them in one catalogue entry defeats the purpose.

Open-weight models broke the assumption that one model belongs to one provider. The same weights can run on several clouds and in several countries, with different prices and residency consequences.

For a compliance reviewer, these are separate items. A model served from Japan, the US, or mainland China makes three different propositions, and only one may suit a particular workload.

The catalogue should therefore be organised by deployment rather than lab. Each entry names the model, serving infrastructure, operating region, and jurisdiction whose law applies there. It also names the operating partner when a foreign cloud runs domestically through a local company.

A catalogue organised as 'provider, model, price' cannot answer the buyer's actual question. One organised as 'model, endpoint, region, jurisdiction, price' can answer it from the same table.

It also makes coverage gaps visible. When a model has no endpoint in a required jurisdiction, teams can plan around that fact in the catalogue instead of finding it through a failed production request.



01

More notes

Next in the set